agent-os

Operate and configure AgentOS itself: the `agentos` CLI, agentos.toml, gateway/Web UI, providers and models, skills, channels, sessions, cron, sandbox, and memory. Use when: (1) the user asks to change an AgentOS setting (model, provider, router tier, auth, channels, search), (2) starting, stopping, or debugging the gateway or Web UI, (3) installing, updating, or removing skills and taps, (4) inspecting sessions, usage/cost, cron jobs, or diagnostics, (5) migrating from another agent runtime. NOT for: authoring new skills (see docs/features/skills.md), operating other agent CLIs, or modifying AgentOS source code.

Operate Robinhood Agentic Trading through the Robinhood Trading MCP for account and portfolio analysis, market research, order previews, order placement or cancellation, rebalancing, and bounded trading automation. Use when the user mentions Robinhood Agentic accounts, Robinhood Trading MCP, portfolio or buying-power checks, Robinhood orders, an agentic trading strategy, or publishing content based on Robinhood trading activity.

Use when the user asks to remember, recall, forget, update, search, or inspect durable AgentOS memory, including profile facts in USER.md and long-term notes in MEMORY.md or memory/**/*.md.

Look up Robinhood tokenized-stock (RWA) contract addresses and tickers. Use when: the user asks for the Robinhood token, on-chain contract address, ticker/symbol, or chain of a real-world stock or ETF (e.g. 'what is Apple's ticker', 'mã cổ phiếu Apple là gì', 'Robinhood contract address for Tesla', 'Microsoft token address'). Resolves a company name OR ticker to the on-chain token on Robinhood Chain. NOT for: live stock prices, trade execution, or non-Robinhood tokens. No API key needed.

Render HTML (with CSS) to a PDF file. Trigger when the user wants to export a styled report, invoice, label, or any HTML/Jinja-rendered page to PDF. Uses WeasyPrint, which supports a meaningful subset of CSS Paged Media (page size, margins, headers/footers, page-break-before/after). Optional dependency — install via `pip install 'use-agent-os[document-extras]'` or `uv add weasyprint` because WeasyPrint pulls in native libraries (Pango, Cairo, fontconfig) that need OS-level packages.

Query the web through multiple search engines (Brave, Tavily, SerpAPI, DuckDuckGo, Bing, Baidu, Sogou, 360) with a single CLI surface. Trigger when the user asks for a research search, fact lookup, source discovery, or wants to compare engines for coverage. The skill aggregates per-engine result lists and normalizes them into a uniform JSON shape for downstream skills (deep-research is the primary consumer). API-key engines gate themselves on the relevant environment variable; engines requiring no key always run.

Submit audio or video for multilingual dubbing, poll status, and download dubbed audio. Use when the user asks for dubbing, 多语言配音, 视频翻译配音, 译制片, or wants a source clip dubbed into another language.

Generate a structured short-video shooting script from a topic. Emits a strict, machine-parseable shot list (3 shots by default) with image prompt + video prompt + voiceover + on-screen text per shot. Trigger when the user asks for a video script, 分镜, 短视频文案, AI视频, 短剧脚本, or wants visual prompts ready for image/video generation.

Use when the user asks to schedule recurring tasks, one-off reminders, timers, or cron-style jobs through the AgentOS cron tool.

Multi-round research with explicit methodology, evidence tracking, and citation-tagged synthesis. Trigger on 'deep dive', 'research report', 'literature review', 'investigate X across sources', 'multi-round investigation'. Distinct from the `summarize` skill, which is a single-pass condensation; this skill maintains a state file across iterations, tracks coverage, and produces a long-form report with per-claim citations. Three execution stages: plan (scope into sub-questions), iterate (record evidence per round), compile (synthesize report). The skill itself does not fetch the web — it tells the host agent which fetches to perform via AgentOS's existing web tools, and records what comes back.

Read, edit, or create Microsoft Word `.docx` files. Trigger this skill whenever the user mentions a Word document, .docx file, contract, report, brief, memo, or asks to extract text, modify an existing doc, generate one from a brief, or audit tracked changes. Three execution paths: text-and-structure extraction, in-place edit-by-run (preserves styles), and create-from-scratch with python-docx. Falls back to OOXML unzip-and-patch for layout work python-docx cannot reach.

Capture the current git diff (staged, working-tree, or staged file list) as text. Direct shell call for workflows that need repository diffs without an LLM agent loop.

GitHub operations via `gh` CLI: issues, PRs, CI runs, code review, API queries. Use when: (1) checking PR status or CI, (2) creating/commenting on issues, (3) listing/filtering PRs or issues, (4) viewing run logs. NOT for: complex web UI interactions requiring manual browser flows (use browser tooling when available), bulk operations across many repos (script with gh api), or when gh auth is not configured.

Query the per-turn DecisionEntry log for skill co-occurrence patterns and the router fixture corpus. Returns a JSON summary suitable for downstream LLM consumption. Useful for 'which skills did I use most this week?'

Fetch a URL via HTTP/HTTPS and return the response body as text. Lightweight entrypoint replacement for `sub-agent` steps whose only job is a single GET/POST. Supports GET (default), POST/PUT/DELETE with a stdin-piped body, configurable timeout, and a max-bytes cap — no LLM agent loop, no custom-header injection (request goes out with urllib defaults). Use for simple data-fetch steps in meta-skill DAGs; for crawling, JS-rendered pages, or complex auth chains use sub-agent + scrapling instead.

Generate instrumental music, background beds, jingles, or sung songs with lyrics through AgentOS audio tools. Use when the user asks for BGM, music generation, 唱歌, 生成歌曲, lyrics to song, or a playable music audio artifact.

Generate or edit a single image via OpenRouter (google/gemini-3.1-flash-image-preview by default). Accepts a text prompt and optional --input-image for image-to-image editing. Trigger when the user asks for an AI image, illustration, concept art, product render, or wants to modify an existing image.

Edit PDFs with natural-language instructions using the nano-pdf CLI.

Structured `.pdf` operations: extract text/tables, merge pages from multiple PDFs, split a PDF by page ranges, fill PDF form fields, and generate fresh PDFs from JSON. Trigger when the user wants programmatic PDF work without natural-language rewriting — examples: pull tables from a report, combine three PDFs, extract pages 5-12, fill a tax form, or build a new PDF from data. Distinct from `nano-pdf`, which uses an LLM to rewrite a page from a sentence; this skill is deterministic byte-level work via pypdf, pdfplumber, and reportlab.

Read, edit, or create PowerPoint .pptx files. Trigger this skill whenever the user mentions a deck, slides, slide deck, presentation, or a `.pptx` filename — whether the goal is to extract text, modify an existing deck, build one from scratch, or prepare slides for review. Three execution paths are supported: text extraction (always available), template editing (unzip → patch slide XML → repack), and creation from scratch (python-pptx for Python or PptxGenJS for Node).

Render a single 3-15s video clip via Seedance 2.0. Supports two backends: OpenRouter (default, model bytedance/seedance-2.0) and the official Volcengine ARK / BytePlus ModelArk endpoint (model doubao-seedance-2-0-260128 / dreamina-seedance-2-0-260128). Accepts a structured English video prompt, optional first-frame image, and optional identity/style reference image. Trigger when the user asks for AI video clip generation, 分镜视频, seedance, or wants a short cinematic shot from a prompt + frame.

Build an SRT subtitle file from a 3-shot short-drama script (ai-video-script OUTPUT FORMAT). Reads each SHOT_N block's DURATION_S + VOICEOVER, emits cumulative-timestamped SRT cues. Pure text-processing, no LLM, no network. Used by meta-short-drama between merge and the final subtitle-burn step.

Internal helper for meta-stack-trace-investigator. Use when a stack trace language is unknown and the workflow needs language-neutral failure-contract checks, reproducer guidance, and patch targets.

Internal helper for meta-stack-trace-investigator. Use when a Go panic or stack trace needs Go-specific nil/error checks, go test reproducer guidance, and patch targets.

Internal helper for meta-stack-trace-investigator. Use when a JavaScript or TypeScript stack trace needs npm/node/tsc-specific checks, reproduction guidance, and patch targets.

Internal helper for meta-stack-trace-investigator. Use when a Python traceback needs Python-specific root-cause checks, pytest reproducer guidance, and defensive patch targets.

Internal helper for meta-stack-trace-investigator. Use when a Rust panic or backtrace needs Rust-specific Result/Option checks, cargo test guidance, and patch targets.

Delegate a self-contained task to a sub-Agent (Codex, Claude Code, or Pi via background process). The original use case was coding tasks — building features, reviewing PRs, refactoring — but the skill is the generic "spawn a sub-Agent with full tool surface" slot used by meta-skill DAG steps for any LLM-driven sub-task (policy review, trace parsing, report synthesis, document generation). Renamed from ``coding-agent`` to reflect actual usage; the wrapped CLIs (codex / claude / pi) still bias toward coding workloads. Use when: (1) building/creating new features or apps, (2) reviewing PRs (spawn in temp dir), (3) refactoring large codebases, (4) iterative tasks that need file exploration, (5) meta-skill steps requiring full tool/LLM agency. NOT for: simple one-liner fixes (just edit), reading code (use read tool), thread-bound ACP harness requests in chat (for example spawn/run Codex or Claude Code in a Discord thread; use sessions_spawn with runtime:"acp"), or any work in ~/clawd workspace (never spawn agents

Burn an SRT subtitle file into an MP4 via ffmpeg's subtitles filter (libass). Single-pass re-encode of video; audio copied as-is. CJK-friendly font fallback chain (Microsoft YaHei → SimHei → Arial Unicode MS → Arial). Used by meta-short-drama as the final subtitling step after merge.

Summarize, condense, or digest content

Read a UTF-8 text file and emit its raw content on stdout. Tiny helper for meta-skills that need to round-trip an artefact through disk so the user can hand-edit it between steps (e.g. tweak script.txt during a review pause). Unlike the builtin read_file tool — which returns line-numbered output for model display — this returns bytes verbatim, suitable for downstream parsers.

Render a static title / ending card PNG with Pillow. Centered headline + optional subtitle on a solid-colour background. CJK-friendly font fallback (Microsoft YaHei → SimHei → Songti → Noto CJK → bitmap). Pure deterministic, no LLM, no network. Used by meta-short-drama for opening and closing cards.

Remote-control tmux sessions for interactive CLIs by sending keystrokes and scraping pane output. Use when: (1) monitoring Claude/Codex sessions running in tmux, (2) sending input to interactive terminal applications, (3) scraping output from long-running processes inside tmux, (4) navigating tmux panes/windows programmatically, or (5) checking on background work in existing sessions. NOT for: one-off shell commands (use exec_command), starting new background processes (use background_process), or non-tmux interactive processes.

Concatenate a directory of numbered MP4 segments (1_*.mp4, 2_*.mp4, ...) into one MP4 with optional fade transitions, unified resolution/fps/codec. Pure ffmpeg wrapper, no LLM. Trigger when a workflow has produced several short clips that need stitching into a final reel.

Turn a single still image (PNG/JPG) into a short MP4 with a slow Ken-Burns zoom and a silent audio track. Pure ffmpeg wrapper. Designed as the on_failure substitute for AI video-gen steps that get blocked by content moderation: when seedance refuses, this skill emits a valid replacement clip from the already-generated still so a downstream merge can still produce a complete deliverable.

Create and register cloned voices for later TTS only when the speaker has explicit consent. Use when the user asks for voice clone, clone voice, 克隆音色, 复刻声音, or wants a reusable voice_id.

Convert a local source recording into an authorized target voice. Use when the user asks for voice conversion, voice changer, 换声, 变声, 音色转换, or converting existing narration to another approved voice.

Generate narration, product voiceover, IVR prompts, podcast reads, or short-video VOICEOVER audio through AgentOS audio tools. Use when the user asks for TTS, 配音, 旁白, 口播, audio narration, or wants script text turned into a playable audio artifact.

Get current weather and forecasts via wttr.in or Open-Meteo. Use when: user asks about weather, temperature, or forecasts for any location. NOT for: historical weather data, severe weather alerts, or detailed meteorological analysis. No API key needed.

Read, edit, or create Microsoft Excel `.xlsx` workbooks. Trigger this skill whenever the user mentions a spreadsheet, .xlsx file, workbook, sheet, formula, pivot table, or asks to extract tabular data, modify a sheet, or build a workbook from rows. Three execution paths: structured inspection, in-place cell edits, and create-from-scratch via openpyxl. Values starting with `=` are written as formulas; everything else is a literal value with type preserved (int / float / str / datetime).